Magnetoencephalography (MEG) and Electroencephalography (EEG) inverse problem is well-known to require regularization to avoid ill-posedness. Usually, regularization is based on mathematical criteria (minum norm, ...). Physiologically, the brain is organized in functional parcels and imposing a certain homogeneity of the activity within these parcels was proven to be an efficient way to analyze the MEG/EEG data [1][6]. The parcels information can be computed from diffusion Magnetic Resonances Imaging (dMRI) by grouping together source positions shared the same connectivity profile (computed as tractograms from diffusion images). In this work, three parcel-based inverse problem approaches have been tested. The first two approaches are based on minimum norm with added regularization terms to account for the parcel information. They differ by the use of a hard/soft constraint in the way they impose that the activity is constant within each parcel [4]. The third approach is based on the Maximum Entropy on Mean (MEM) framework [2]. The dMRI-base and random cortex parcellation, we test also the use of Multivariate Source Pre-localization (MSP) [5] in the source reconstruction.